Dynamical brain states can be characterized by non-linear measures of EEG. The present study shows that critical transitions, i.e., abrupt changes from one dynamic pattern of neural mass activity to another one, may be detected by abrupt variations in local chaoticity. Using an ambulatory device, EEG was recorded from 10 patients with a schizophrenic and two patients with an affective disorder during a series of 25-min interviews. Dynamical aspects, in particular, phase transitions in the EEG-dynamics of the EEG were characterized by means of a measure that continuously estimates the chaoticity of the EEG signal and is thus related to its predictability. Results indicate simpler dynamics of the EEG time series in paranoid-hallucinatory patients, while at the same time these patients tended to exhibit more abrupt transitions/unit of time between different dynamical EEG states. Such sudden phase transitions in brain activity were significantly enhanced prior to expressions of thought disorders that were detected by the interviewer and an observer in the conversation, compared with time periods during the interview without such symptoms.